Near-solidus Melting of the Shallow Upper Mantle: Partial Melting Experiments on Depleted Peridotite

Transcription

1 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 PAGES 1163± Near-solidus Melting of the Shallow Upper Mantle: Partial Melting Experiments on Depleted Peridotite LAURA E. WASYLENKI*, MICHAEL B. BAKER, ADAM J. R. KENT AND EDWARD M. STOLPER DIVISION OF GEOLOGICAL AND PLANETARY SCIENCES, CALIFORNIA INSTITUTE OF TECHNOLOGY, MAIL CODE , PASADENA, CA 91125, USA RECEIVED SEPTEMBER 18, 2001; ACCEPTED FEBRUARY 5, 2003 We present the results of melting experiments on a moderately depleted peridotite composition (DMM1) at 10 kbar and 1250±1390 C. Specially designed experiments demonstrate that liquids extracted into aggregates of vitreous carbon spheres maintained chemical contact with the bulk charge down to melt fractions of 002±004 and approached equilibrium closely. With increasing melt fraction, SiO 2, FeO*, and MgO contents of the partial melts increase, Al 2 O 3 and Na 2 O contents decrease, and CaO contents first increase up to clinopyroxeneout at a melt fraction of 009±010, then decrease with further melting. A linear fit to melt fraction vs temperature data for lherzolite-bearing experiments yields a solidus of C. The melting reaction is 056 orthopyroxene 072 clinopyroxene 004 spinel ˆ 034 olivine 1 liquid. Above clinopyroxene-out, the reaction is 124 orthopyroxene ˆ 024 olivine 1 liquid. Near the solidus, DMM1 glass compositions have lower SiO 2, TiO 2,Na 2 O, and K 2 O contents, higher FeO*, MgO, and CaO contents, and higher CaO/Al 2 O 3 ratios compared with glasses from low-degree melting of fertile peridotite compositions. Recent computational models predict partial melting trends generally parallel to our experimental results. We present a parameterization of 10 kbar peridotite solidus temperatures suggesting that K 2 O and P 2 O 5 have greater effects on solidus depression than Na 2 O, consistent with theoretical expectations. Our parameterization also suggests that abyssal peridotites have 10 kbar solidi of 1278±1295 C. KEY WORDS: depleted; experimental petrology; mantle melting; near-solidus; peridotite *Corresponding author. Present address: Department of Geological Sciences, 4044 Derring Hall, Virginia Polytechnic Institute and State University, Blacksburg, VA 24061, USA. Telephone: Fax: lew@vt.edu ypresent address: Department of Geosciences, Wilkinson Hall, Oregon State University, Corvallis, OR 97331, USA. INTRODUCTION It is widely accepted that parental magmas of midocean ridge basalts (MORBs) represent a mixture of liquids produced by pressure-release melting over a range of depths in the Earth's upper mantle (e.g. McKenzie, 1984; McKenzie & Bickle, 1988; Langmuir et al., 1992). Experimental and theoretical studies indicate that small amounts of melt (52%) are interconnected in olivine-dominated, partially molten systems (e.g. Waff & Bulau, 1979; von Bargen & Waff, 1986; Daines & Richter, 1988) and that such low degrees of melt can move relative to the residual solid phases (McKenzie, 1985, 1989; Stevenson & Scott, 1991). The extremely depleted rare earth element patterns observed in residual clinopyroxenes from dredged abyssal peridotites have been used as evidence of this mobility and that melt production beneath mid-ocean ridges can approach the limit of fractional melting ( Johnson et al., 1990; Johnson & Dick, 1992; Salters & Dick, 2002). Correlations between the major element compositions of spinels and the concentrations of moderately incompatible trace elements in clinopyroxenes from abyssal peridotites also suggest that these peridotites experienced near-fractional melting (Hellebrand et al., 2000). Finally, melt inclusions with variably depleted trace element concentrations relative to average MORB (e.g. Sobolev & Shimizu, 1993; Sobolev, 1996; Shimizu, 1998) suggest that erupted MORB magmas contain a component of liquid produced by melting of depleted peridotite. Taken together, these lines of evidence suggest that Journal of Petrology 44(7) # Oxford University Press 2003; all rights reserved.

2 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 JULY 2003 partial melts can be extracted from the mantle beneath mid-ocean ridges at low melt fractions and that at least some components of MORBs were produced by melting of depleted sources with low concentrations of incompatible major, minor, and trace elements relative to the fertile MORB source. Most high-pressure experimental studies of peridotite melting have focused on fertile mantle compositions, and these studies have provided information on melt compositions over a wide range of pressures and temperatures (e.g. Jaques & Green, 1980; Takahashi, 1986; Hirose & Kushiro, 1993; Baker & Stolper, 1994; Baker et al., 1995; Kushiro, 1996; Robinson et al., 1998; Walter, 1998). In addition, Pickering-Witter & Johnston (2000) and Schwab & Johnston (2001) explored the effects of pyroxene and spinel abundances and pyroxene compositions on liquid compositions, melt productivities, and melting reactions at 10 kbar, and, by varying the orthopyroxene/clinopyroxene ratios in their starting materials, they were able to work on bulk compositions spanning wide ranges of MgO, Al 2 O 3, CaO, and Na 2 O contents. In a synthesis of available experimental data, Hirschmann et al. (1998a) showed the important role that alkalis play in controlling the compositions of olivine orthopyroxene clinopyroxene-saturated melts of peridotite. Likewise, the broadly inverse correlation between total alkali contents of peridotites and their solidus temperatures (Herzberg et al., 2000; Hirschmann, 2000) suggests that, at a given pressure, depleted peridotites will begin to melt at higher temperatures than fertile peridotites. Because peridotites in the shallowest part of the melting zone beneath midocean ridges are expected to have been depleted in incompatible major and minor elements (and especially alkalis) by prior melting, the expectation is that such peridotites would melt to a lower degree and produce different liquids compared with melting of fertile peridotite under similar conditions. In this paper we present the results of melting experiments on a moderately depleted peridotite composition (DMM1) at 10 kbar and 1250±1390 C; these results complement the large amount of available data on more fertile peridotite compositions. We used a variant of the diamond aggregate melt-extraction technique developed by Johnson & Kushiro (1992), Hirose & Kushiro (1993), and Baker & Stolper (1994) to study near-solidus experimental glasses unmodified by quench crystal growth. In the experiments described here, the aggregate comprised vitreous carbon spheres or fragments (Wasylenki et al., 1996; Pickering-Witter & Johnston, 2000; Schwab & Johnston, 2001). A potential problem associated with the diamond aggregate technique in low melt fraction experiments is that pressure is initially low within the void spaces of the aggregate until the layer is completely filled with melt (Baker et al., 1996), possibly leading to the segregation of liquid that is far from equilibrium with the peridotite at the actual pressure of the experiment. The motivation for replacing the porous diamond aggregate used in our previous studies (Baker & Stolper, 1994; Baker et al., 1995) with vitreous carbon spheres is that, although vitreous carbon is strong enough to support open pore space at the start of an experiment, it is much less stiff than diamond (Noda et al., 1969; Sawa & Tanaka, 2002), so this open space collapses more quickly. Another advantage is that vitreous carbon is readily polished at the end of an experiment, allowing small pools of melt to be imaged and analyzed in situ with ease (e.g. Pickering-Witter & Johnston, 2000; Schwab & Johnston, 2001). In addition to modifying the melt-extraction technique in this way, we have addressed the controversy surrounding this technique (see Baker et al., 1996; Falloon et al., 1996, 1997, 1999; Wasylenki et al., 1996; Pickering-Witter & Johnston, 2000) with special experiments that demonstrate the reliability and close approach to equilibrium of meltextraction experiments. Following the presentation of the experimental results, we consider the effects of bulk composition on 10 kbar melting reaction coefficients and on melt productivity, and we compare our experimental glass compositions for the depleted peridotite DMM1 with those produced by melting more fertile peridotite compositions at the same pressure. We also compare our experimental glass compositions with liquid compositions calculated for the DMM1 composition using four peridotite melting models. Finally, by combining our experimentally determined 10 kbar solidus temperature with solidus determinations on other peridotite compositions, we develop a simple expression for predicting the solidus temperature at 10 kbar as a function of bulk composition. EXPERIMENTAL AND ANALYTICAL TECHNIQUES Starting materials Starting materials for this study were prepared by mixing mineral separates (olivine, ol; orthopyroxene, opx; clinopyroxene, cpx; spinel, sp) from a Kilbourne Hole spinel lherzolite nodule (KBH), olivine from a Hawaiian dunite nodule (H1801i), and synthetic diopside (CaMgSi 2 O 6 ). Electron microprobe analyses of the natural minerals are reported in Table 1. The crystallinity of the synthetic diopside was verified by X-ray diffraction, and the composition reported in Table 1 is that of stoichiometric diopside. The natural minerals were hand-picked from disaggregated 1164

3 WASYLENKI et al. PARTIAL MELTING OF DEPLETED PERIDOTITE Table 1: Compositions of starting materials and bulk composition of DMM1 SiO 2 TiO 2 Al 2 O 3 Cr 2 O 3 FeO* MnO MgO CaO Na 2 O K 2 O NiO mg-no. KBH ol(10) (16) ÐÐ 0.01(1) 0.01(1) 9.34(10) 0.12(2) 49.49(21) 0.07(1) ÐÐ ÐÐ 0.39(2) 90.4(5) H1801i ol(14) (17) ÐÐ 0.01(1) 0.02(2) 12.07(71) 0.17(2) 47.55(55) 0.16(2) ÐÐ ÐÐ 0.39(4) 87.5(15) KBH opx(8) 54.99(22) 0.10(2) 4.76(15) 0.50(4) 5.98(6) 0.13(2) 32.79(14) 0.86(2) 0.12(1) ÐÐ 0.11(2) 90.7(5) KBH cpx(15) 51.99(21) 0.41(5) 6.56(11) 1.00(5) 2.81(4) 0.09(2) 15.24(10) 19.89(11) 1.58(3) ÐÐ 0.04(2) 90.6(8) Diopside ÐÐ ÐÐ ÐÐ ÐÐ ÐÐ ÐÐ ÐÐ ÐÐ ÐÐ KBH sp(11) 0.11(2) 0.11(2) 54.91(27) 12.96(20) 10.54(11) 0.10(2) 20.95(10) ÐÐ ÐÐ ÐÐ 0.37(2) 78.0(5) DMM (11) 0.04(1) 2.38(5) 0.39(1) 8.34(15) 0.13(1) 41.59(15) 2.14(2) 0.055(3) 0.006(2) 0.28(1) 89.9(5) FB 46.59(8) 0.41(2) 13.78(5) 0.32(1) 12.07(10) 0.10(2) 11.95(3) 12.06(3) 1.39(2) ÐÐ 0.01(1) 63.8(3) GBR ÐÐ ÐÐ 18.2 ol, olivine; opx, orthopyroxene; cpx, clinopyroxene; sp, spinel; DMM1, bulk composition used in this study; FB, ferrobasalt; GBR, Glass Buttes rhyolite. FeO*, all Fe as FeO; mg-no. ˆ 100MgO/(MgO FeO*) on a molar basis. 1 Minerals designated by KBH were separated from a Kilbourne Hole nodule; number of electron microprobe analyses in parentheses, each analysis on a separate grain. Numbers in parentheses adjacent to each oxide value are one sample standard deviation in terms of the least units cited, e.g (16) represents Dashes indicate that the element was not analyzed or was below the detection limit of the microprobe. 2 Olivine separated from a dunite nodule collected from the Hualalai 1801 flow. 3 Synthetic crystalline diopside; crystallinity verified by X-ray diffraction. Listed composition is for end-member diopside. 4 Bulk composition (with the exception of K 2 O and P 2 O 5 ) calculated from constituent mineral compositions and mineral proportions given in text. Mean bulk K 2 O content calculated from liquid K 2 O contents in runs 22, 12, 27C, 28C, 33C, and 42C; phase proportions; and K 2 O partition coefficients from Halliday et al. (1995). Mean bulk P 2 O 5 ˆ (12); calculated from liquid P 2 O 5 contents in runs 28C, 33C, and 42C (Table 3); phase proportions; and ol±liq, opx±liq, and cpx±liq partition coeffcients for P 2 O 5 of 0.1, 0.03, and 0.05 (Libourel et al., 1994; Brunet & Chazot, 2001). Uncertainties (with the exception of those for K 2 O and P 2 O 5 ) are based on Monte Carlo propagations of errors on phase compositions and estimated weighing errors; uncertainty on K 2 O and P 2 O 5 is one sample standard deviation of the mean bulk K 2 O and P 2 O 5 values. 5 Composition of Glass Buttes rhyolite taken from Dobson et al. (1989) and Ihinger et al. (1999). nodules, ground, sieved to 16±28 mesh, cleaned in warm 24 N hydrochloric acid, and rinsed in deionized water. They were then crushed further, sieved to 200± 325 mesh, and again washed in 24 N HCl and deionized water. Finally, the grains were ground and sieved to 13 mm. The synthetic diopside was ground by hand until most of the grains were 515 mm, but the powder still contained a few grains as large as 40 mm in size. The minerals were mixed in the weight proportions 0432 KBH olivine, 0202 H1801i olivine, 0274 KBH orthopyroxene, 0014 KBH clinopyroxene, 0060 diopside, and 0018 KBH spinel to generate the depleted peridotite starting material (DMM1). The mixture was ground by hand for 1 h under ethanol to ensure homogeneity. The bulk composition of DMM1 in terms of major oxides is reported in Table 1. With the exceptions of K 2 O and P 2 O 5, these values were computed from the mineral compositions and their proportions in the mix (see preceding paragraph). K 2 O and P 2 O 5 contents for DMM1 were calculated using the K 2 O and P 2 O 5 contents of glasses in selected experiments, mineral±melt partition coefficients from the literature, and melt fractions and residual mineral proportions for these experiments derived from mass balance constraints using the other oxides (see footnotes to Tables 1 and 3). The starting DMM1 composition was chosen to represent a moderately depleted mantle peridotite. When compared with the residues in the 10 kbar melting experiments of Baker & Stolper (1994) and Baker et al. (1995), DMM1 is roughly equivalent to the residue formed by 12±13% batch melting of MM3. Figure 1 compares the concentrations of MgO, Al 2 O 3, CaO, and Na 2 O in DMM1 with other peridotites studied experimentally at 9±10 kbar, with estimates of the primitive mantle, and with reconstructed abyssal peridotites. DMM1 has lower Al 2 O 3, CaO, and Na 2 O contents than all estimates of primitive mantle, and, unlike other peridotite compositions that have been studied experimentally, it falls on the welldefined MgO±oxide trends for abyssal peridotites (Fig. 1). Figure 1a and b shows that DMM1 is lower in Al 2 O 3 than all other experimentally studied peridotite compositions and lower in CaO than the others, except those investigated by Sen (1982; 5 in Fig. 1b) and by Pickering-Witter & Johnston (2000; 9 in Fig. 1b; note that this composition is far removed from the MgO±CaO trend defined by abyssal peridotites and primitive mantle estimates). Figure 1c shows that DMM1 also has lower Na 2 O than most peridotite compositions previously studied. 1165

4 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 JULY 2003 Fig. 1. MgO±oxide variation diagrams showing the bulk composition of the moderately depleted DMM1 peridotite used in this study (large filled square) and estimates of primitive upper-mantle compositions (*: Ringwood, 1979; Sun, 1982; Wanke et al., 1984; Hart & Zindler, 1986; Allegre et al., 1995; McDonough & Sun, 1995), reconstructed abyssal peridotite compositions (^: Baker & Beckett, 1999), and other peridotitic starting materials studied experimentally at 9±10 kbar (*: 1, MM3, Baker & Stolper, 1994; 2, PHN1611, Kushiro, 1996; 3, KLB-1, 4, HK-66, Hirose & Kushiro, 1993; 5, 77PAII-1, Sen, 1982; 6, Hawaiian Pyrolite±40% olivine (HPy-40); 7, Tinaquillo Lherzolite±40% olivine, Jaques & Green, 1980; 8, FER-B, 9, FER-C, 10, FER-D, 11, FER-E, Pickering-Witter & Johnston, 2000; 12, INT-A, 13, INT-B, 14, INT-D, 15, INT-E, Schwab & Johnston, 2001). MgO vs (a) Al 2 O 3, (b) CaO, (c) Na 2 O. Experimental methods All experiments were run in a 127 cm piston cylinder apparatus using CaF 2 cells, straight-walled graphite furnaces, and inner pieces of crushable MgO dried at 1000 C for at least 8 h. Pressure was applied using the hot-piston-in technique with no friction correction. Experiments at 1300 C with identical assemblies bracketed the Ca-Tschermak breakdown reaction to lie at 11±14 kbar. This pressure range encompasses the reaction boundary (1300 C, 13 kbar) determined by Hays (1966). W 97 Re 3 /W 75 Re 25 thermocouples were used to monitor and control temperature to within 1 C of the set point. No pressure correction was applied to the nominal e.m.f.±temperature relation. Based on past experiments with double thermocouples, temperatures are estimated to be accurate to within 15 C. Run durations and experimental conditions are reported in Table 2. Except in experiments 1, 2, and 3, N 2 gas was bled into the slot in the thermocouple plate during each experiment to minimize oxidation of the thermocouple wires within and just below the steel base plug. At the end of each experiment, the thermocouple wires just below the base plug (i.e. within the run assembly) were inspected for signs of oxidation; no evidence of significant oxidation was observed on any of the wires, including those from experiments 1±3. After the power was turned off, samples cooled to below 1000 C within several seconds. As each run assembly was taken apart, the position of the capsule relative to the center of the furnace and that of the thermocouple junction relative to the top of the capsule were measured to ensure that the capsule was properly positioned within the furnace and that the thermocouple tip was 05±1 mm from the top of the capsule. Each capsule was sliced vertically with a diamond wafering blade, mounted in epoxy, and polished for electron microprobe analysis. This study includes three types of melting experiments. The first type of experiment consisted of two stages (see Baker & Stolper, 1994). For the first stage, 3±6 mg of peridotite powder that had been dried in a vacuum oven at 110 C for 2 h was loaded into a graphite inner capsule. The inner capsule was then placed in a 015 inch o.d. Pt capsule. The composite capsule assembly was then dried (see Table 2); and finally the crimped end of the Pt capsule was welded. After drying and welding, the capsule was run at temperature and pressure for 52±136 h. For the second stage, the silicate charge was removed from the firststage capsule and loaded into a new graphite capsule along with 80±100 mm diameter vitreous carbon spheres that had been dried for at least 1 h in a 110 C vacuum oven. The mass of vitreous carbon (Table 2) was 3±9% of the mass of the silicate sample. The loaded second-stage graphite capsule was then placed in a Pt capsule, after which the composite assembly was dried and welded shut. This second-stage capsule was then run for 18±135 h at the same temperature and 1166

5 WASYLENKI et al. PARTIAL MELTING OF DEPLETED PERIDOTITE Table 2: Experimental results Run Temp. ( C) Drying 1st stage (h) 2nd stage (h) Silicate (mg) Glassy carbon (mg)y Phases present Phase proportions (wt %)z /4,300/ gl, ol, opx 16.6(6), 68.1(1.5), 15.4(1.6) /1,400/ gl, ol, opx 13.6(5), 68.1(1.7), 18.3(1.7) /12,110/ gl, ol, opx 12.5(6), 67.5(1.5), 20.0(1.6) /20,300/ gl, ol, opx 10.0(5), 67.1(1.5), 22.9(1.6) 20C / (33.8, FB) gl, ol, opx 11.7(4), 66.9(1.5), 21.4(1.5) 26C / (9.8, GBR) gl, ol, opx 12.3(5), 66.6(1.4), 21.1(1.5) 25T 1285/ / / gl, ol, opx 10.8(5), 66.0(1.5), 23.2(1.7) 23T 1310/ /1 25.5/ gl, ol, opx, cpx 8.8(5), 66.5(7), 23.5(8), 1.2(4) /3,110/ gl, ol, opx 10.3(5), 66.8(1.5), 22.9(1.7) /1,110/ gl, ol, opx, cpx, sp 6.6(8), 65.0(1.6), 24.0(1.7), 4.2(7), 0.3(2) 17T 1285/ /1 68.4/ gl, ol, opx, cpx, sp 3.2(8), 65.5(1.5), 25.3(1.7), 5.8(7), 0.3(1) /1,300/ gl, ol, opx, cpx, sp 6.4(7), 66.4(1.4), 23.5(1.6), 3.4(7), 0.3(1) 27C / (21.6, FB) gl, ol, opx, cpx, sp 5.4(8), 64.1(1.8), 26.3(2.0), 3.7(7), 0.5(2) 28C / (26.9, FB) gl, ol, opx, cpx, sp 2.6(7), 64.5(1.5), 26.8(1.6), 5.8(8), 0.4(2) 33C / (22.0, FB) gl, ol, opx, cpx, sp 1.6(3), 64.7(1.3), 26.5(1.4), 6.8(5), 0.5(2) 34C / (40.6, GBR) gl, ol, opx, cpx, sp 4.0(8), 64.7(1.4), 26.0(1.6), 5.0(7), 0.3(2) 42C / (31.8, FB) gl, ol, opx, cpx, sp 3.3(6), 62.6(1.6), 27.9(1.7), 5.4(5), 0.8(2) /4,400/ ol, opx, cpx, sp 62.8(1.7), 28.4(1.9), 8.1(5), 0.6(2)x Phases present: gl, glass; ol, olivine; opx, orthopyroxene; cpx, clinopyroxene; sp, spinel. Indicates drying procedure for loaded capsule before welding. Temperature and duration in hours (rounded to nearest hour; 12 indicates that the capsule was dried overnight) separated by slash. Drying conditions for both stages given for two-stage runs. ynumbers in parentheses indicate masses, in micrograms, of aggregate chips included with silicate samples in compositionalconvergence runs. FB and GBR indicate glass used in aggregate chip; FB, synthetic ferrobasaltic glass; GBR, Glass Buttes rhyolite (compositions listed in Table 1). zcalculated by mass balance. Numbers in parentheses indicate uncertainties; e.g. 16.6(6) represents % liquid. The silica contents of 17 of the 18 olivine compositions used to calculate the modes have been adjusted using the stoichiometric constraint for ol that Si (cation %) equals For each of these analyses, the SiO 2 (wt %) value was iteratively raised or lowered and cation % calculated until the Si value was within units of All oxides were then normalized to 100 wt %; it should be noted that this procedure has no effect on the Mg/(Mg Fe) ratio of the olivines. Although the correction is relatively small (e.g. mean absolute value of the change in silica contents is 0.26 wt %), the correlation coefficients for melt fraction (F) vs ol and F vs opx increased dramatically (e.g. for F vs ol, the Pearson correlation coefficient increased from 0.05 to 0.58). Unadjusted olivine compositions are reported in Table 3. xcalculated using analyses for opx and cpx that were normalized to 100 wt % (see Table 3). pressure as its corresponding first-stage run. At the end of each second-stage experiment, the charge contained quenched glass associated with the vitreous carbon. The glass formed rinds generally 3±15 mm thick around individual vitreous carbon spheres. Varying amounts of glass and crushed carbon spheres filled the spaces between the intact spheres (Fig. 2a). The idea behind this type of experiment was to avoid the possibility of formation and segregation of disequilibrium liquids at the start of the experiment that would not subsequently equilibrate with the main mass of the sample. In twostage experiments, liquids maintain intimate contact with the peridotite and approach equilibrium during the first stage and are segregated only in the second stage. The second type of experiment was run as a single stage, but the temperature was changed during the experiment to evaluate whether liquid that had segregated into the interstices between the vitreous carbon spheres could respond chemically to changes in temperature; these experiments are indicated by a T next to the run number in Table 2. Graphite capsules were loaded with vitreous carbon and peridotite powder, placed in Pt outer capsules, and then dried and welded shut as for the previously described charges. Each sample was run at 10 kbar and an initial temperature as listed in Table 2 for 26±68 h. Following this initial period, the temperature was raised either 15 or 40 C, and the charge was held at this higher temperature for an additional 88±117 h. The potential concern 1167

6 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 JULY 2003 Fig. 2. Backscattered electron photomicrographs of (a) a typical charge with vitreous carbon spheres and (b) a portion of the charge from run 28C, showing the chip of vitreous carbon aggregate that had been pre-impregnated with ferrobasalt before the experiment. addressed by these experiments is whether liquid segregated into the vitreous carbon at the start of the experiment is in sufficient chemical contact with the main mass of the experimental charge that its composition can change by diffusive interaction with the peridotite and its interstitial liquid. We refer to the third type of experiment as `compositional-convergence' runs; these experiments are marked with a C next to the run number in Table 2. For these seven experiments, glass-bearing vitreous carbon aggregates were prepared as follows. An oxide mix of a synthetic ferrobasalt or a powdered rhyolitic glass from Glass Buttes, Oregon (both compositions are reported in Table 1), were mixed with conchoidally fractured vitreous carbon shards (80±147 mm in size). Approximately 85% (by weight) of the material in each capsule was vitreous carbon. Each of these two mixtures was loaded into a separate Pt capsule and then held at 10 kbar and 1350 C for a few hours. At the end of these two preparatory runs, the silicate liquids had impregnated most of the pore spaces between the vitreous carbon fragments. Fragments (10±41 mg; see Table 2) of the glass-impregnated vitreous carbon aggregates were loaded into graphite±pt capsules with DMM1 powder (3±6 mg; see Table 2) as described above for the other two types of experiments. The samples were run for 127±188 h. In one experiment (20C), a layer of vitreous carbon spheres was also loaded into the capsule at the opposite end of the charge from the ferrobasalt-impregnated vitreous carbon aggregate. The purpose of these experiments was to examine the degree to which the ferrobasaltic and rhyolitic glasses in the interstices of the vitreous carbon aggregate were able to converge in composition toward each other and toward the compositions of glasses within vitreous carbon aggregates that were not pre-filled with glass (i.e. glass within the layer of vitreous carbon in 20C or that in charges from the previously described experiments). Because the ferrobasaltic and rhyolitic glasses are extremely distant in composition from the equilibrium partial melts of the DMM1 peridotite, the extent to which they can shift compositionally during an experiment can be used to demonstrate that liquids segregated into vitreous carbon aggregates can maintain chemical contact with the main mass of the peridotite during an experiment. This means that even if a disequilibrium liquid migrates into the vitreous carbon aggregate at the start of an experiment, it can evolve substantially via exchange with the bulk charge over the course of the experiment. Moreover, if the initial ferrobasaltic and rhyolitic compositions converge to a single composition similar to glasses from our two-stage and temperaturechange experiments, we can infer with confidence that an equilibrium liquid composition was approached closely. Oxygen fugacity was not controlled in our experiments, but the presence of graphite inner capsules constrains f O2 to below the graphite±co vapor buffer (GCO). Following the approach discussed by Bertka & Holloway (1988) and Gudmundsson et al. (1988), we placed 13 mg of platinum wire in the middle of the peridotite powder in experiment 25 (1325 C). During the experiment the Pt wire absorbed Fe. Based on the compositions of the resulting Fe±Pt alloy and coexisting orthopyroxene and olivine, and using the equations and solution models of Jamieson et al. (1992), the log 10 f O2 in run 25 was ±92, or 24 log units below the quartz±fayalite±magnetite buffer (QFM) at 10 kbar (Huebner, 1971). The slightly different activity± composition models of Gudmundsson & Holloway (1993) yielded a log 10 f O2 of 85, or 17 log units below QFM. The inferred f O2 values of our experiments are thus consistent with estimates for the MORB source region (Christie et al., 1986; Green et al., 1987; O'Neill & Wall, 1987; Mattioli & Wood, 1988; Mattioli et al., 1989). 1168

7 WASYLENKI et al. PARTIAL MELTING OF DEPLETED PERIDOTITE Analytical techniques Experimental charges were analyzed at Caltech with a five spectrometer JEOL 733 electron microprobe using an accelerating voltage of 15 kev. Crystalline phases in all experiments (including the Pt alloy in run 25) were analyzed with a beam current of 30 na and a rastered beam at magnifications greater than (the resulting spot size is less than 2 mm 2 mm). Glasses were analyzed with a 25, 5, or 10 na beam current and a rastered area as large as possible for the glass pools being analyzed (generally 5 mm 5 mm). Repeated analyses on the same region of glass within several experiments indicated little or no Na loss. All data were processed using CITZAF (Armstrong, 1988). Two basaltic glasses, VG-2 and BGIO ( Jarosewich et al., 1979), were analyzed as secondary standards during each microprobe session to estimate accuracy and precision of the instrument on a long-term basis; when crystalline phases were analyzed, Johnstown hypersthene and Natural Bridge diopside ( Jarosewich et al., 1979) were also analyzed to provide additional estimates of accuracy and precision. Raw oxide sums for the experimental glasses were typically 95±97%, but were occasionally lower, and in one case down to 89%. We consistently observed higher totals for glass pools that were more than 5 mm in the shortest dimension and lower totals for smaller glass pools. The low totals probably reflect the inclusion of vitreous carbon in the electron beam analysis volume. This inference is supported by the observation that when all of the glass analyses for a given sample are normalized to 100%, there are no systematic differences in the normalized oxide concentrations as functions of the raw oxide sums. Within the vitreous carbon aggregate of a given experiment, normalized glass compositions display no systematic variation as a function of distance from the DMM1 peridotite. Mean compositions for normalized glasses, crystalline phases, and secondary standards are reported in Table 3. Water contents of the glasses in two charges (3 and 22) were measured by secondary ion mass spectrometry with the modified Cameca IMS-3f ion microprobe at Lawrence Livermore National Laboratories using the techniques described by Kent et al. (1999). Measured H / 30 Si ratios in our experimental glasses were converted to water contents with a calibration curve constructed by analyzing a set of Marianas back-arc basin basaltic glasses (Stolper & Newman, 1994) and synthetic MORB glasses (Dixon et al., 1995) whose water and silica contents were determined by independent techniques. Errors on the water contents of the experimental glasses are conservatively estimated at 15% relative. Charges 3 and 22 contain and wt % water, respectively. These have not been corrected for the possible effects of vitreous carbon in the analysis volume. As vitreous carbon can dissolve hydrogen in the presence of water-bearing melt, but contains no Si (L. Wasylenki, unpublished data, 1995), such a correction would probably lead to lower estimates of the water contents of our experimental glasses. RESULTS Evaluation of the approach to equilibrium Although the melt-extraction technique used in this study provides a simple way to avoid the quench modification problems encountered in conventional peridotite melting experiments close to the solidus, the technique must be applied with care. In particular, the issue of whether the glasses within the vitreous carbon aggregates (this study; Pickering-Witter & Johnston, 2000; Schwab & Johnston, 2001) or diamond aggregates (e.g. Johnson & Kushiro, 1992; Hirose & Kushiro, 1993; Baker & Stolper, 1994; Baker et al., 1995; Kushiro, 1996) actually represent equilibrium melt compositions has sparked much debate (Baker et al., 1996; Falloon et al., 1996, 1997, 1999; Wasylenki et al., 1996). For example, as explained above, one possible problem with the one-stage experiments of Baker & Stolper (1994) is that at the onset of an experiment, low pore pressure within the diamond layer could affect the composition of liquid initially filling these pores, i.e. the initial liquid may not be in equilibrium with the bulk peridotite at the nominal pressure of the experiment. Similarly, the kinetics of melting could lead to generation of a disequilibrium liquid at the start of an experiment. In either of these cases, liquid migrating rapidly into the pores in the diamond or vitreous carbon layer might not be in equilibrium with the bulk of the material in the capsule, and, if this segregated liquid were subsequently cut off from exchange with the adjacent peridotite, results would be erroneous. Liquids segregated into diamond aggregates in short-duration, one-stage experiments do indeed differ from those produced in long-duration runs at the same temperature and pressure (Johnson & Kushiro, 1992; Baker & Stolper, 1994), so this general class of potential problems cannot be disregarded. Our two-stage experiments, however, were designed to minimize these problems in that the liquid moving into a vitreous carbon layer at the beginning of a second-stage run is remelted glass from the first-stage run and thus should represent a near-equilibrium melt. Temperature-change experiments Although our two-stage experiments were explicitly designed to minimize problems of melt re-equilibration 1169

8 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 JULY 2003 Table 3: Experimental results Run Phase SiO 2 TiO 2 Al 2 O 3 Cr 2 O 3 FeO* MnO MgO CaO Na 2 O K 2 O Sum 9 gl(6) 50.96(26) 0.17(11) 12.19(11) 0.38(6) 8.36(18) 0.19(7) 16.58(22) 10.65(24) 0.44(3) 0.08(1) 100 ol(5) 41.36(18) 0.01(1) 0.05(1) 0.25(2) 8.94(9) 0.13(2) 49.67(26) 0.24(2) ÐÐ ÐÐ opx(12) 56.70(24) 0.02(2) 1.83(21) 0.81(6) 5.26(13) 0.11(2) 33.75(21) 1.30(14) 0.02(1) ÐÐ gl(5) 50.38(37) 0.25(8) 12.87(15) 0.42(7) 7.93(31) 0.22(6) 15.43(25) 12.18(19) 0.30(3) 0.02(1) 100 ol(8) 40.96(15) ÐÐ 0.05(1) 0.27(3) 8.96(12) 0.13(2) 49.15(38) 0.27(1) ÐÐ ÐÐ opx(11) 55.73(20) 0.03(2) 3.00(26) 0.84(8) 5.62(8) 0.12(2) 32.22(26) 1.82(6) 0.02(1) ÐÐ gl(6) 50.21(21) 0.21(4) 14.15(13) 0.36(5) 7.72(17) 0.15(4) 14.35(17) 12.23(19) 0.55(3) 0.07(1) 100 ol(6) 40.92(15) ÐÐ 0.04(1) 0.24(2) 8.96(9) 0.14(3) 49.23(26) 0.24(2) ÐÐ ÐÐ opx(8) 55.39(42) 0.02(1) 2.94(28) 0.94(5) 5.73(20) 0.10(2) 32.55(34) 1.93(17) 0.02(1) ÐÐ gl(9) 49.45(21) 0.28(8) 15.43(14) 0.27(5) 7.34(16) 0.16(7) 12.77(20) 13.61(22) 0.60(4) 0.08(1) ol(3) 41.30(15) ÐÐ 0.04(1) 0.21(2) 9.24(9) 0.12(2) 49.26(26) 0.31(1) ÐÐ ÐÐ opx(4) 54.86(38) 0.04(2) 3.58(25) 0.98(6) 5.68(18) 0.10(2) 31.93(34) 2.45(7) 0.02(1) ÐÐ C gl in vc(4) 49.50(24) 0.33(8) 14.92(13) 0.31(5) 7.34(16) 0.10(3) 13.49(16) 13.23(21) 0.67(5) 0.10(1) 100 gl in ic(4) 49.38(21) 0.32(9) 15.15(14) 0.30(5) 7.37(16) 0.11(4) 13.32(16) 13.25(21) 0.70(1) 0.09(1) 100 ol(9) 40.88(18) ÐÐ 0.05(1) 0.21(2) 9.21(10) 0.13(2) 48.98(26) 0.30(1) ÐÐ ÐÐ opx(6) 56.12(20) 0.03(1) 2.53(16) 0.78(5) 5.69(9) 0.13(2) 32.97(18) 2.18(6) 0.02(1) ÐÐ C gl(6) 50.19(28) 0.27(4) 13.78(14) 0.27(4) 7.81(21) 0.15(7) 14.10(33) 12.81(20) 0.56(3) 0.07(2) 100 ol(8) 40.62(15) ÐÐ 0.06(2) 0.16(1) 9.32(11) 0.14(3) 49.36(30) 0.27(1) ÐÐ ÐÐ opx(7) 55.95(22) 0.02(2) 2.56(26) 0.91(7) 5.79(8) 0.10(2) 32.79(29) 1.81(9) 0.02(1) ÐÐ T gl(7) 49.21(24) 0.30(4) 14.16(14) 0.23(3) 7.61(17) 0.11(4) 14.18(17) 13.55(21) 0.56(3) 0.07(1) 100 ol(4) 41.11(29) ÐÐ 0.05(2) 0.19(2) 9.13(19) 0.13(2) 49.24(29) 0.30(1) ÐÐ ÐÐ opx(4) 55.94(20) 0.02(2) 2.78(30) 0.74(9) 5.66(14) 0.13(2) 32.59(18) 2.10(9) 0.02(1) ÐÐ T gl(7) 48.99(24) 0.28(9) 15.37(17) 0.22(6) 7.91(17) 0.09(5) 12.60(29) 13.66(22) 0.81(10) 0.07(1) 100 ol(8) 40.84(15) ÐÐ 0.05(1) 0.17(3) 9.44(10) 0.14(3) 49.60(26) 0.29(3) ÐÐ ÐÐ opx(6) 54.86(20) 0.05(2) 3.65(22) 0.98(5) 5.82(7) 0.12(2) 32.07(27) 2.32(7) 0.03(1) ÐÐ cpx(6) 52.10(28) 0.03(2) 5.10(19) 1.34(8) 3.80(11) 0.11(2) 20.02(26) 17.04(30) 0.12(2) ÐÐ gl(11) 49.64(21) 0.28(4) 15.12(14) 0.29(4) 7.41(16) 0.13(3) 13.32(16) 13.12(21) 0.60(3) 0.08(1) 100 ol(9) 41.13(15) ÐÐ 0.04(1) 0.19(4) 9.25(11) 0.13(2) 49.07(26) 0.28(1) ÐÐ ÐÐ opx(2) 55.30(52) 0.05(1) 3.24(10) 0.79(4) 5.63(15) 0.10(2) 31.69(17) 2.54(13) 0.03(1) ÐÐ gl(7) 48.08(51) 0.39(6) 17.73(20) 0.17(3) 7.07(16) 0.18(5) 12.02(30) 13.32(21) 0.96(5) 0.08(1) 100 ol(5) 39.58(21) ÐÐ 0.07(4) 0.18(5) 9.45(9) 0.14(3) 49.67(26) 0.27(2) ÐÐ ÐÐ opx(4) 55.22(45) 0.01(1) 2.44(23) 0.83(10) 5.85(7) 0.10(2) 33.35(18) 1.68(13) 0.02(1) ÐÐ cpx(4) 50.89(30) 0.12(3) 6.27(36) 1.00(5) 3.91(12) 0.10(2) 19.95(38) 16.82(22) 0.14(2) ÐÐ sp(2) 0.29(2) 0.11(2) 34.19(32) 35.95(54) 10.35(10) 0.14(3) 18.66(10) 0.02(1) 0.01(1) ÐÐ T gl(11) 48.81(35) 0.39(18) 16.89(15) 0.19(7) 7.28(24) 0.14(9) 12.14(14) 12.85(20) 1.14(5) 0.17(2) 100 ol(10) 40.81(15) ÐÐ 0.05(1) 0.13(1) 9.60(14) 0.14(3) 48.71(26) 0.27(1) ÐÐ ÐÐ opx(5) 54.24(54) 0.08(2) 4.83(46) 0.79(7) 5.84(7) 0.13(2) 31.57(39) 2.18(11) 0.05(2) ÐÐ cpx(4) 52.07(39) 0.11(3) 5.47(16) 1.02(7) 3.83(13) 0.11(2) 20.03(38) 17.39(38) 0.21(3) ÐÐ sp(3) 0.60(52) 0.09(2) 47.62(43) 21.89(23) 8.92(9) 0.11(2) 20.35(37) 0.05(1) ÐÐ ÐÐ gl(11) 48.88(20) 0.41(10) 17.61(16) 0.19(5) 6.84(15) 0.16(5) 11.72(16) 13.21(21) 0.84(4) 0.14(2) 100 ol(8) 40.83(15) ÐÐ 0.03(1) 0.11(1) 9.42(13) 0.14(3) 49.15(26) 0.25(1) ÐÐ ÐÐ opx(4) 55.30(20) 0.04(2) 3.49(26) 0.92(9) 5.78(7) 0.12(2) 32.27(39) 2.25(21) 0.02(1) ÐÐ cpx(4) 52.46(34) 0.08(2) 5.23(51) 1.15(5) 3.80(8) 0.12(2) 20.00(36) 17.80(41) 0.15(2) ÐÐ sp(2) 0.26(7) 0.06(3) 44.60(95) 25.1(1.1) 9.32(42) 0.12(4) 20.06(11) 0.02(1) ÐÐ ÐÐ C gl(13) 48.22(20) 0.48(7) 15.95(16) 0.12(4) 7.30(21) 0.09(6) 12.75(26) 13.89(22) 1.12(6) 0.09(2) 100 ol(5) 40.82(30) ÐÐ 0.05(1) 0.08(1) 9.56(9) 0.13(2) 49.07(44) 0.24(1) ÐÐ ÐÐ opx(8) 55.02(48) 0.05(1) 3.70(14) 0.88(15) 5.98(11) 0.12(2) 31.66(17) 2.21(10) 0.04(2) ÐÐ

9 WASYLENKI et al. PARTIAL MELTING OF DEPLETED PERIDOTITE Run Phase SiO 2 TiO 2 Al 2 O 3 Cr 2 O 3 FeO* MnO MgO CaO Na 2 O K 2 O Sum cpx(10) 52.17(46) 0.09(2) 5.15(54) 1.20(12) 3.82(9) 0.10(2) 18.72(24) 18.04(33) 0.18(3) ÐÐ sp(2) 0.39(32) 0.09(2) 39.8(1.5) 30.1(2.1) 10.51(17) 0.13(2) 19.73(60) 0.04(1) ÐÐ ÐÐ C gl(6) 48.07(20) 0.58(25) 15.92(25) 0.15(6) 7.54(18) 0.15(6) 12.35(19) 13.72(22) 1.38(10) 0.14(2) 100 ol(4) 41.01(15) ÐÐ 0.07(1) 0.09(3) 9.65(13) 0.13(2) 48.44(29) 0.30(3) ÐÐ ÐÐ opx(6) 53.76(27) 0.07(2) 5.01(36) 0.80(5) 6.05(11) 0.12(2) 31.35(17) 2.26(10) 0.04(2) ÐÐ cpx(4) 51.72(37) 0.11(5) 5.61(42) 1.01(5) 3.88(15) 0.11(2) 19.07(65) 17.60(65) 0.28(3) ÐÐ sp(4) 0.84(17) 0.10(2) 48.8(1.0) 19.82(87) 9.29(17) 0.12(2) 20.90(15) 0.07(2) ÐÐ ÐÐ C gl(10) 47.96(20) 0.63(8) 17.77(16) 0.09(3) 7.29(16) 0.13(4) 10.35(12) 13.19(21) 2.08(10) 0.51(1) 100 ol(6) 40.99(15) ÐÐ 0.05(2) 0.08(2) 9.61(15) 0.12(3) 48.42(26) 0.24(1) ÐÐ ÐÐ opx(6) 54.28(20) 0.08(2) 5.04(22) 0.73(5) 5.99(7) 0.13(2) 31.19(17) 2.28(11) 0.04(1) ÐÐ cpx(4) 51.21(24) 0.16(3) 5.90(38) 0.97(5) 3.90(12) 0.10(2) 19.08(25) 17.84(14) 0.27(2) ÐÐ sp(2) 0.75(61) 0.12(3) 49.34(56) 19.2(1.1) 9.66(35) 0.14(5) 20.4(1.0) 0.07(4) ÐÐ ÐÐ C gl(7) 48.83(20) 0.50(7) 16.87(15) 0.11(3) 7.19(16) 0.18(5) 11.51(22) 13.04(21) 1.43(7) 0.33(2) 100 ol(4) 40.83(15) ÐÐ 0.04(1) 0.10(2) 9.72(20) 0.14(3) 48.08(26) 0.25(2) ÐÐ ÐÐ opx(8) 54.27(20) 0.04(1) 4.58(29) 0.88(12) 5.97(7) 0.12(2) 31.02(25) 2.18(9) 0.04(2) ÐÐ cpx(3) 51.75(54) 0.11(6) 4.74(24) 1.16(18) 3.89(9) 0.12(2) 19.30(30) 17.60(38) 0.31(14) ÐÐ sp(3) 0.82(42) 0.08(2) 48.2(1.5) 20.8(1.7) 9.58(9) 0.12(2) 19.92(18) 0.06(3) ÐÐ ÐÐ C gl(7) 47.93(20) 0.59(7) 17.54(16) 0.11(3) 7.28(17) 0.13(4) 11.42(17) 13.25(21) 1.60(8) 0.14(2) 100 ol(9) 40.59(15) ÐÐ 0.05(1) 0.08(2) 9.74(15) 0.13(2) 49.44(26) 0.23(1) ÐÐ ÐÐ opx(4) 54.17(40) 0.06(2) 3.93(12) 0.93(21) 5.91(16) 0.13(2) 32.45(50) 2.19(6) 0.04(2) ÐÐ cpx(6) 51.83(19) 0.09(2) 4.87(37) 1.21(5) 3.76(9) 0.11(2) 19.71(40) 18.07(25) 0.17(3) ÐÐ sp(3) 0.49(23) 0.11(2) 42.4(1.7) 28.5(1.8) 10.19(13) 0.14(3) 19.73(24) 0.05(3) ÐÐ ÐÐ ol(9) 40.84(18) ÐÐ 0.07(5) 0.06(2) 9.70(11) 0.13(2) 49.07(30) 0.14(4) ÐÐ ÐÐ 100 opx(7) (65) 0.07(2) 4.80(25) 0.71(8) 6.09(15) 0.13(2) 32.17(52) 1.95(11) 0.05(2) ÐÐ 100 cpx(16) (40) 0.19(6) 5.74(32) 0.98(8) 3.67(15) 0.10(2) 18.67(60) 18.52(33) 0.42(15) ÐÐ 100 sp(3) 1.21(1.03) 0.10(1) 47.74(74) 20.37(93) 10.24(21) 0.13(2) 19.30(24) 0.11(14) 0.01(1) ÐÐ VG2(176) (43) 1.80(12) 13.93(20) 0.02(3) 11.55(21) 0.20(4) 6.89(11) 10.88(15) 2.79(8) 0.21(2) BGIO(167) 51.30(49) 1.28(10) 15.19(24) 0.05(3) 8.98(20) 0.16(4) 8.04(14) 11.14(20) 2.78(14) 0.09(2) JHYP(103) 54.46(46) 0.08(2) 0.85(4) 0.78(4) 14.72(25) 0.48(3) 27.33(22) 1.16(5) 0.01(1) 0.01(1) DIOP(116) 55.60(52) 0.02(2) 0.20(10) 0.02(2) 0.29(3) 0.03(2) 18.06(21) 25.69(37) 0.17(6) 0.01(1) C, compositional-convergence experiment; T, temperature-change experiment (see text for further discussion); gl, glass; ol, olivine; opx, orthopyroxene; cpx, clinopyroxene; sp, spinel; vc, layer of vitreous carbon spheres; ic, impregnated vitreous carbon chip; FeO*, all Fe as FeO. Numbers in parentheses after each phase are the number of analyses averaged for that phase; numbers in parentheses after each oxide value are 1s of the sample distribution in terms of the least units cited, e.g (26) represents Glass compositions normalized to 100% on a volatile-free basis. Measured H 2 O contents for glasses from runs 3 and 22 are 1.2(2) and 0.3(1) wt %, respectively. Measured P 2 O 5 contents in glasses from runs 28C, 33C, and 42C are 0.11(5), 0.16(1), and 0.12(2) wt %, respectively. Dashes indicate that the element was not analyzed or was below the detection limit. 1 Reported analyses have been normalized to 100 wt %. Repeated attempts to analyze the silicate phases in this charge consistently produced low analytical totals. The opx and cpx compositions reported here have nominal sums of 98.6 and 98.9 wt %, respectively. Following the calculations outlined by Cameron & Papike (1981), both compositions have acceptable pyroxene stoichiometries. 2 Secondary standards. VG2, basaltic glass, Juan de Fuca Ridge; BGIO, basaltic glass, Indian Ocean; DIOP, diopside, Natural Bridge; JHYP, hypersthene, Johnstown meteorite (Jarosewich et al., 1979). Mean oxide values for VG2 and BGIO analyzed during glass-analysis and mineral-analysis microprobe sessions, respectively, overlap at the 1s level. Thus, all VG2 analyses have been averaged together as have all BGIO analyses. in the vitreous carbon layers, we nevertheless made further efforts to demonstrate directly that liquids in the vitreous carbon layers in our experiments can change composition over the course of an experiment. The temperature-change experiments described above provide evidence of continued interaction between segregated melt and the peridotite over the course of an experiment. We conducted three temperature-change experiments to demonstrate that segregated liquids can change composition in response to changes in experimental conditions. In each such experiment, after 1171

10 JOURNAL OF PETROLOGY VOLUME 44 NUMBER 7 JULY 2003 initial run durations of 26±68 h at 1285 or 1310 C, temperatures were raised either 15 or 40 C and the runs continued for an additional 88±117 h. We assume that during the initial stages of these experiments, liquids in the vitreous carbon layers approached the glass compositional trends defined by the two-stage experiments at the same initial temperatures. In two of the three temperature-change experiments (17T, 1285±1300 C; 25T, 1285±1325 C), the final liquid compositions are similar to those in two-stage experiments run at the same final temperatures and different from liquid compositions expected at 1285 C (see Fig. 4). (The third experiment, 23T, appears to have had a significant temperature gradient, as evidenced by modal variations from top to bottom of the charge, and to have had a final temperature of 1310 C rather than 1325 C; see the section `Phase relations and liquid compositions' below.) Based on mass balance calculations using the quenched glass and mineral compositions and the bulk DMM1 composition, the melt fractions of 17T and 25T during their respective high-temperature steps were 003 and 011 (Table 2). Thus, over the course of 93±117 h at these melt fractions, the segregated liquids in the vitreous carbon layers were able to change in response to the increase in temperature via diffusive exchange with liquid retained in the peridotite. We also attempted temperature-change experiments in which the temperature was lowered in the second phase of the experiment, but, as one would expect, this resulted in extensive crystallization within the vitreous carbon aggregates. Compositional-convergence experiments Successful `compositional-convergence' experiments demonstrate not only that liquids in vitreous carbon aggregates can change significantly, but also by convergence that an equilibrium result has been approached from more than one direction in multicomponent composition space. As described above, these experiments involved placing 10±41 mg of a vitreous carbon aggregate that was pre-impregnated with either a synthetic ferrobasalt or a rhyolite in a capsule with DMM1 powder (Fig. 2b). Based on the results of our two-stage experiments, both the ferrobasalt and the rhyolite glass compositions were initially far from equilibrium with DMM1 at 10 kbar and any temperature. With the exceptions of K 2 O and, to a lesser extent, Na 2 O for the two rhyolite-bearing experiments, the few micrograms of glass added to the system have little effect on the bulk composition. The potassium budget in the two rhyolite-bearing runs (26C and 34C) is dominated by the rhyolite glass; for example, we estimate that the bulk K 2 O content in 34C increased from 0006 wt % (the estimated bulk composition of DMM1) to 002 wt % as a result of the addition of the rhyolitebearing vitreous carbon aggregate. The change in the sodium content in run 34C is less extreme; the Na 2 O concentration increased from 006 to 007 wt %. Figure 3 illustrates the large shifts in composition that the ferrobasaltic and rhyolitic glasses underwent as a result of diffusive equilibration with 2±12 wt % melt in the peridotite and shows that the quenched liquids in all of these experiments define a single compositional trend approached from opposite directions in most compositional dimensions. As shown in Fig. 4, this compositional trend is consistent with the trend defined by glass compositions from the two-stage and temperature-change experiments. It is important to emphasize that the compositions of the liquids in the vitreous carbon chips in these experiments have converged to their final compositions from opposite directions in terms of several oxides and that the compositional changes achieved in these experiments are in some cases very large (e.g. SiO 2, CaO, MgO, and FeO* for the initially rhyolitic liquid). It is also striking that three of the convergence experiments are within 55 C of our estimated solidus for the DMM1 composition and have only 2±5% melt (Table 2). These results should thus dispel controversy about the validity of melt-segregation experiments into diamond or vitreous carbon aggregates, as they demonstrate large compositional shifts, convergence from melts initially in the aggregates that are extremely far in composition from the final compositions, and that the results of two-stage, temperaturechange, and `compositional-convergence' experiments form coherent compositional trends as a function of melt fraction near the solidus. Homogeneity of crystalline phases We have also examined the compositions of the residual peridotite minerals in all of our experiments (Table 3), as true achievement of equilibrium would require homogeneous crystals and mineral compositions that are independent of distance from the glassbearing aggregates. In all experiments olivine grains (15±40 mm in size) were homogeneous within analytical error in each charge, both from core to rim of individual analyzed grains and from top to bottom of the residual peridotite. Orthopyroxene grains with cross-sectional diameters larger than 10 mm often displayed small, incompletely reacted cores, visible with high-gain, backscattered electron imaging. We estimate that these cores occupy 535% of the volume of opx in a given charge. The cores have compositions between the initial KBH opx 1172

11 WASYLENKI et al. PARTIAL MELTING OF DEPLETED PERIDOTITE Fig. 3. Changes in melt compositions in `compositional-convergence' experiments (see text): (a) SiO 2 vs Al 2 O 3 ; (b) FeO* (all Fe as FeO) vs Na 2 O; (c) MgO vs CaO. The large gray square in each panel represents the initial composition of ferrobasaltic glass in aggregates of vitreous carbon particles that were loaded in capsules with DMM1 powder and run at 10 kbar for 127±184 h. Smaller open and filled squares represent compositions of glasses in the vitreous carbon chips at the end of the experiments. Glass composition from the vitreous carbon layer in run 20C is also plotted (^). The large gray circle in each panel represents the initial composition of Glass Buttes rhyolite in another set of vitreous carbon aggregates run with DMM1 powder at 10 kbar for 166±188 h; smaller open and filled circles represent the compositions of the glasses at the end of these experiments. Glass compositions denoted by small filled squares or circles coexist with spinel lherzolite; small open squares or circles denote harzburgite residues. Arrows show typical compositional changes in these `compositional-convergence' experiments; no arrows are drawn from the initial ferrobasalt composition to the re-equilibrated melts in (c) because of the close proximity of the two sets of symbols. Melt fractions (F) are given for the experiments that define each end of the compositional trends. composition and analyzed rim compositions, indicating some compositional change toward equilibrium. We did not observe any significant differences in opx core and rim compositions as a function of distance from the vitreous carbon aggregates in any of our experiments. Clinopyroxene grains are typically 515 mm in size. Unreacted cpx cores were not clearly visible in backscattered electron images, but microprobe analyses indicate that cpx grains in the run products are chemically heterogeneous. In particular, CaO and Al 2 O 3 contents vary within and among the cpx grains in each charge. In our study, more than 20 cpx analyses were collected from each charge, and those that clustered toward the low-al and low-ca end of the trend in Al 2 O 3 ±CaO±MgO space defined by all analyses within a charge were averaged for Table 3. None of the cpx analyses from any of the run products were close in composition to the starting KBH cpx composition or pure diopside, suggesting that all of the cpx in our experiments at least partially re-equilibrated. We did not observe any systematic variations in cpx compositions as a function of distance from the vitreous carbon aggregates, but in some experiments modal cpx abundance increases with increasing distance from the top of the capsule (i.e. the end closest to the thermocouple junction). The experiments in which this modal variation occurs probably experienced larger thermal gradients than those experiments that show a more uniform cpx distribution (Lesher & Walker, 1988). Spinel is a minor phase in our experiments and, as run temperatures fall, shows more compositional heterogeneity within runs than olivine or pyroxenes. Spinel grains in the run products are small (510 mm in size), and generally only 3±6 grains are visible on the polished surface of each charge. Thus we were not able to assess the degree of compositional variation within individual sp grains or as a function of the position of sp grains in the charge relative to the melt-bearing vitreous carbon aggregates. We did observe that in a given charge, grains of sp in direct contact with melt tended to be more compositionally similar than those sp grains that were not obviously in contact with melt. The compositions reported in Table 3 are averages of the sp grains that had similar compositions within each charge. As we discuss below, these mean compositions are consistent with the compositional trends of sp from other 9±12 kbar peridotite melting experiments. Overall, although the pyroxenes and spinels in the experimental charges are not homogeneous within analytical error, our run durations were long compared with most previous peridotite melting experiments, and the observed heterogeneity is attributable to incompletely reacted cores that would be difficult to react fully without much finer starting materials or much longer experiments. As these cores do not make up a major fraction of the residual phases, it is unlikely that they have a significant influence on the results. More importantly, the compositional heterogeneities 1173